I have a (rather different!) application that’s released as a Flatpak, and GPU acceleration is CUDA-only there, too. It supports ROCm when compiled locally, but ROCm just can’t work through the sandbox at this point, unfortunately. Not for lack of trying.
If you have an example of a Flatpak where it does work, I’d love to see their manifest so I can learn from it.
I did buy a (secondhand) nvidia card specifically for AI worlkloads because yes, I realised that this is what the AI dev community has settled on, and if I try to avoid nvidia I will be making life very hard for myself.
But that doesn’t change the fact that it still absolutely sucks that nvidia have this dominance in the space, and that it is largely due to what tooling the community has decided to use, rather than any unique hardware capability which nvidia have.
Huh? I can’t speak to “video processing”, but “NVIDIA dominance” isn’t applicable at all for AI, at least generative AI. Pretty much all frameworks for LLM either officially support AMD or have an AMD fork. And literally all image gen frameworks that I am aware of officially support AMD.
I typically see people recommending to buy an AMD GPU over NVIDIA in AI communities…
What AI workloads do you run?
EDIT: Even the app in the OP only supports NVIDIA GPUs for the prepackaged flathub installation. You can run this software with an AMD GPU.
Seems to only support cpu or cuda, not AMD sadly.
https://github.com/ladaapp/lada
I have a (rather different!) application that’s released as a Flatpak, and GPU acceleration is CUDA-only there, too. It supports ROCm when compiled locally, but ROCm just can’t work through the sandbox at this point, unfortunately. Not for lack of trying.
If you have an example of a Flatpak where it does work, I’d love to see their manifest so I can learn from it.
@PlantPowerPhysicist @HappyFrog f(l)atpak is workaround, not solution. It should not work in non-default configuration
For the prepackaged flathub or docker installations.
Great news, you can run this software with an AMD GPU fine, just follow the installation instructions in your link and install pytorch for ROCm. https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/3rd-party/pytorch-install.html
That’s pretty much the standard for a lot of video processing applications
I did buy a (secondhand) nvidia card specifically for AI worlkloads because yes, I realised that this is what the AI dev community has settled on, and if I try to avoid nvidia I will be making life very hard for myself.
But that doesn’t change the fact that it still absolutely sucks that nvidia have this dominance in the space, and that it is largely due to what tooling the community has decided to use, rather than any unique hardware capability which nvidia have.
Huh? I can’t speak to “video processing”, but “NVIDIA dominance” isn’t applicable at all for AI, at least generative AI. Pretty much all frameworks for LLM either officially support AMD or have an AMD fork. And literally all image gen frameworks that I am aware of officially support AMD.
I typically see people recommending to buy an AMD GPU over NVIDIA in AI communities…
What AI workloads do you run?
EDIT: Even the app in the OP only supports NVIDIA GPUs for the prepackaged flathub installation. You can run this software with an AMD GPU.